PAI official images

Python

Overview

Python is the most widely used programming language in machine learning and has a rich library ecosystem that includes NumPy, PyTorch, and TensorFlow. PAI provides two types of Python images:

• CPU images: Built on the official Ubuntu base image for CPU computing.

• GPU images: Built on the official CUDA base image for GPU computing.

Key features

• Supports Ubuntu 22.04 and Ubuntu 24.04.

• Supports Alibaba Cloud's high-performance RDMA networking.

• Supports Python versions 3.10 through 3.14.

• Supports CUDA versions 12.4 through 13.0.

• Includes common development tools such as curl, git, wget, rclone, and ping.

• Uses Alibaba Cloud mirrors for pip and apt.

PyTorch

Overview

PAI provides two types of PyTorch images:

• PAI-based images: Built on PAI Python images with PyTorch, TorchVision, and TorchAudio packages pre-installed. These images inherit all the features of the Python images and cover official PyTorch releases from version 2.4.0 onward.

• NGC-based images: Built on NVIDIA NGC PyTorch images, with common development tools pre-installed and Alibaba Cloud mirrors configured for pip and apt.

Tag descriptions

-accl:

• These images are pre-installed with the Alibaba Cloud High-Performance Collective Communication Library (ACCL). ACCL delivers higher communication performance than NCCL.

• When you develop or train with ACCL-based images, use the preconfigured Python environment. If you want to use a Python virtual environment, follow the installation guide to configure ACCL in your environment.

  -ngc:

• These images are built from NVIDIA NGC PyTorch images. The tag includes the NGC version. For example, 2.10.0-gpu-py312-cu130-ubuntu24.04-ngc25.11 is based on NGC PyTorch 25.11.

• NGC PyTorch image features are documented in the NVIDIA official documentation.

Data-Juicer

Overview

Data-Juicer is a distributed framework for data cleaning and preprocessing. Built on the distributed capabilities of Ray, it is widely used to improve data quality and fuse multimodal data before Large Language Model (LLM) training. PAI provides two types of Data-Juicer images with a full set of built-in data processors, quality evaluation tools, and visual analytics capabilities:

• CPU images: Built on PAI's CPU base image for large-scale CPU-only tasks such as text processing and data cleaning.

• GPU images: Built on PAI's CUDA base image for GPU-accelerated tasks such as model inference and quality scoring.

Key features

• Supports Ubuntu 22.04.

• Supports Alibaba Cloud RDMA for high-throughput, low-latency distributed data loading and processing.

• Includes a built-in Data-Juicer runtime with a full set of processors, so you can quickly start data processing tasks and analyze and monitor them in the Ray Dashboard.

• Supports CPU/GPU heterogeneous resource scheduling for diverse multimodal data workloads such as data cleaning, quality evaluation, and data generation.

• Uses Alibaba Cloud mirrors for pip and apt by default, which improves the speed and stability of dependency installation.

Responsible-AI-Develop

Overview

Responsible AI is a set of principles and practices that spans the entire AI model lifecycle. It ensures that an AI system is safe, fair, transparent, and compliant at every stage. PAI provides two types of base images to support Responsible AI practices:

• CPU images: Built on the official Ubuntu image for general CPU computing, with Responsible AI toolchains integrated.

• GPU images: Built on the official CUDA image for high-performance GPU scenarios, with Responsible AI toolchains integrated.

Key features

• Supports Ubuntu 22.04.

• Supports Python versions 3.11 through 3.14.

• Supports CUDA 11.8.

• Includes Responsible AI visual analytics tools with an interactive dashboard for multidimensional analysis, such as model fairness and error analysis, to help developers identify potential bias and errors.

• Supports differential privacy training, which injects controllable noise during model training to prevent sensitive data leakage and meet data compliance and privacy protection requirements.

• Includes the RAI model encryption SDK (RAI_SAM_SDK) for sharded encrypted storage of LLMs and authorized decryption during inference.

Ray

Overview

Ray is a high-performance framework for distributed computing. It is widely used for large-scale training, hyperparameter search, reinforcement learning, and online inference. PAI provides two types of Ray images with the ray[default] dependency pre-installed, which includes the Ray Dashboard and common runtime components:

• CPU images: Built on PAI's CPU base image for CPU-only distributed computing and data processing.

• GPU images: Built on PAI's CUDA base image for GPU-accelerated training, inference, and large-scale parallel computing.

Key features

• Supports Ubuntu 22.04 and Ubuntu 24.04.

• Supports Alibaba Cloud RDMA for high-throughput, low-latency distributed communication.

• Includes a full Ray runtime environment with common components that allow you to quickly start Ray Head and Worker nodes and run tasks.

• Supports CPU/GPU heterogeneous resource scheduling for diverse workloads such as training, data processing, and inference.

• Uses Alibaba Cloud mirrors for pip and apt by default, which improves the speed and stability of dependency installation.

ModelScope

Overview

ModelScope Library supports the acquisition and management of models and datasets, and model training and inference based on frameworks such as PyTorch and TensorFlow. It is tested on Python 3.8 and later, PyTorch 1.11 and later, and TensorFlow. PAI provides official ModelScope images that work out of the box, with no manual environment setup required. For more information, see ModelScope official images.

TorchEasyRec

Overview

TorchEasyRec is a deep learning framework for recommendation systems. It covers common scenarios such as matching, ranking, multitask learning, and generative recommendation. Simple configuration and convenient custom extensions accelerate the development and rollout of high-performance recommendation models.

PAI provides official TorchEasyRec images pre-installed with dependencies such as pytorch, torchrec, fbgemm, and tensort. Two image types are available:

• GPU version (Recommended): Built on Ubuntu 22.04 with CUDA acceleration for high-performance large-scale recommendation model training.

• CPU version: Built on Ubuntu 22.04 for development, debugging, and small-scale training. Some operations are GPU-only.

CosyVoice-Training

PAI provides official CosyVoice training images. For detailed usage instructions, seeFine-tune CosyVoice 2.0 on DSW.

TensorFlow

The following table lists the framework versions, CUDA versions, and operating systems supported by TensorFlow images.

DeepRec

The following table lists the framework versions, CUDA versions, and operating systems supported by DeepRec images.

XGBoost

The following table lists the framework versions and operating systems supported by XGBoost images.

TritonServer

Overview

Triton Inference Server (Triton Server) is a high-performance inference server developed by NVIDIA. It supports frameworks such as TensorFlow, PyTorch, and ONNX Runtime, and provides a unified interface that simplifies model deployment and inference.

Key features

• Multi-framework support: Triton Server supports various deep learning frameworks and model formats, enabling unified deployment of diverse models.

• High throughput and low latency: Triton improves inference performance through batching and parallel inference. It also uses NVIDIA GPU acceleration to maximize compute power.

• Dynamic model management: Triton allows dynamic loading and unloading of models, which enables flexible version control, A/B testing, and model updates.

• Simple APIs and scalability: Triton offers REST and gRPC interfaces for easy integration. It also integrates with container orchestration systems such as Kubernetes for large-scale inference deployments.

• Heterogeneous hardware support: In addition to NVIDIA GPUs, Triton runs on CPUs and other accelerators, which supports deployment across diverse hardware platforms.

• Custom post-processing: You can apply custom logic to inference results to meet specific application needs.

ComfyUI

Overview

ComfyUI is a node-based graphical interface tool designed for diffusion models such as Stable Diffusion. Using a visual workflow, you drag and drop to build an image generation pipeline and flexibly control the generation logic without writing code.

Key features

• Node-based workflow: Breaks down steps such as text encoding, sampling, model loading, and image post-processing into independent nodes that you can freely connect for precise control.

• Efficient resource management: Loads only the models needed for the current workflow, which reduces VRAM usage and supports batch generation and complex pipeline optimization.

• Highly extensible: Supports custom node plugins with a rich community ecosystem, such as ControlNet, LoRA, and Upscale, for easy integration of new models or features.

• Workflow export and sharing: Entire generation workflows can be exported as JSON files for reproducibility, collaboration, or deployment to other environments.

PAI-RAG

Overview

PAI-RAG is an enterprise-grade retrieval-augmented generation (RAG) conversational system solution from PAI. Built on PAI-EAS, it delivers out-of-the-box RAG capabilities. It combines LLMs with knowledge retrieval so that you can quickly build applications such as private knowledge Q&A and intelligent customer service. It also provides an open source modular framework (GitHub: aigc-apps/PAI-RAG) for flexible customization.

Key features

• Multiple vector database support: Natively compatible with Elasticsearch, Hologres, Tablestore, Milvus, and other mainstream vector databases to meet diverse enterprise needs.

• Web search enhancement: Supports real-time web retrieval to overcome the timeliness limitations of model pretraining data and improve answer accuracy and freshness.

• Flexible deployment and integration: Offers a WebUI, RESTful API, and OpenAI-compatible interface for quick integration into existing business systems.

• Knowledge base full lifecycle management: Supports document upload and management using the WebUI or OSS, with one-stop capabilities for chunking, vectorization, version updates, and knowledge base operations.

vLLM

Overview

vLLM is an open source inference and serving engine for large language models (LLMs), designed to efficiently deploy and run a wide range of open source large models. Its memory management and scheduling techniques significantly increase throughput while keeping latency low, which makes it a mainstream LLM inference framework.

Key features

• PagedAttention: A core innovation inspired by OS paging mechanisms to dynamically manage KV Cache, eliminate VRAM fragmentation, and increase VRAM utilization several times over.

• Continuous batching: Dynamically merges requests of varying lengths for parallel decoding, which greatly improves GPU utilization and throughput.

• High throughput, low latency: Supports higher concurrency on the same hardware and is ideal for high-traffic production environments.

• Developer-friendly: Provides a simple Python API and an OpenAI-compatible interface for rapid integration into existing applications.

• Rich ecosystem: Natively supports advanced features such as LoRA fine-tuning inference, multimodal models, and tool calling (Function Calling).

EasyAnimate

Overview

EasyAnimate is an end-to-end framework for generating high-definition, long-form videos. PAI developed it in-house based on the Diffusion Transformer (DiT) architecture. It quickly generates high-quality videos from text or images (text-to-video and image-to-video) and provides a complete solution that covers data preprocessing, VAE training, and DiT inference.

Key features

• High-resolution long video generation: Generates coherent videos up to 1024×1024 resolution and 6 seconds or longer.

• Multimodal input: Supports both text prompts (text-to-video) and image inputs (image-to-video) for dynamic video generation.

• Complete training pipeline: Offers end-to-end training capabilities for VAE, DiT foundation models, and LoRA fine-tuning to support customized development.

• Production-ready deployment: Officially supported by PAI inference services for seamless integration into cloud inference platforms and is suitable for production environments.

Kohya

Overview

Kohya is an ecosystem toolset derived from Stable Diffusion fine-tuning scripts. The Gradio graphical interface built on top of these scripts lowers the barrier to fine-tuning models with methods such as LoRA and DreamBooth.

Key features

• Multiple training methods: Natively supports LoRA, DreamBooth, full-parameter fine-tuning, and SDXL model training.

• Graphical interface: Provides an intuitive Web UI (based on Gradio) where you can configure parameters using forms instead of command-line coding.

• Cross-platform compatibility: Primarily designed for Windows but also supports Linux and macOS.

• Complete toolchain: Integrates data preprocessing, auto-captioning, training monitoring, and model export to cover the full fine-tuning lifecycle.

• Open source and active community: Fully open source with continuous community maintenance and compatibility with mainstream inference frameworks, such as Stable Diffusion WebUI, for direct deployment of trained models.

Stable-Diffusion-WebUI

Overview

Stable-Diffusion-WebUI is an open source graphical interface tool for deploying and running Stable Diffusion models locally. It lowers the barrier to generative AI tasks such as text-to-image and image-to-image generation.

Key features

• Multimodal generation: Supports mainstream modes such as text-to-image (txt2img), image-to-image (img2img), inpainting, and outpainting.

• Rich extension ecosystem: The built-in plugin system supports popular extensions such as ControlNet, LoRA, and T2I-Adapter to enhance generation control.

• Integrated training and fine-tuning: Includes DreamBooth, LoRA, and Textual Inversion for custom model fine-tuning.

• Cross-platform deployment: Runs on Windows, Linux, macOS, and Google Colab, and is compatible with a range of CPU and GPU (NVIDIA/AMD) hardware environments.

• User-friendly: The web interface built with Gradio offers visual parameter configuration and is suitable for users from beginners to professionals.

CosyVoice

CosyVoice is a next-generation high-fidelity speech synthesis model. It clones a target voice from a prompt audio clip of 30 seconds or less and supports cross-lingual replication. It suits scenarios such as customer service conversations, audiobook narration, and short-video dubbing. CosyVoice images are available in the following two deployment forms.

SGLang

Overview

SGLang (Structured Generation Language) is a high-performance inference and serving framework for large language models. It uses a co-designed frontend language and backend runtime. The frontend provides a structured generation programming language for writing controllable generation logic. The backend is an optimized inference engine, SGLang Runtime, that supports low-latency, high-throughput model serving.

Key features

• Structured controllable generation: Natively supports precise output format control using JSON Schema, regular expressions, and other constraints to overcome the limitations of traditional prompt engineering.

• High-performance inference: Uses innovative optimizations such as RadixAttention and Radix Cache to achieve 3 to 5 times higher throughput than mainstream solutions.

• Multimodal support: Works with both text-only LLMs and vision-language models (VLMs), and supports multimodal inputs such as images and video.

• Flexible integration: Offers a simple Python API that can replace the OpenAI API for complex prompt workflows, which lowers development barriers.

TensorFlow-Serving

Overview

TensorFlow Serving is a high-performance, open source model service system for machine learning models. As a core component of the TensorFlow Extended (TFX) ecosystem, it rapidly deploys trained TensorFlow models in the SavedModel file format as online inference services and exposes them through gRPC and RESTful APIs.

Key features

• Model version management: Supports parallel loading of multiple model versions and seamless rolling updates for phased releases and rollbacks.

• High-performance inference: Production-optimized scheduling and batching mechanisms ensure low-latency, high-throughput service.

• Out-of-the-box integration: Natively supports the TensorFlow SavedModel file format without requiring additional conversion.

• Extensible architecture: Offers pluggable components such as Servable, Source, and Manager for custom loading logic and serving policies.

• Multi-protocol support: Provides both gRPC (high performance) and HTTP/REST (easy integration) interfaces to accommodate different client needs.

Lingjun resource images (Serverless)

The following table lists the core training images for Lingjun resources (Serverless).

AIGC images

The following table lists the core AIGC images. For a description of the Stable-Diffusion-WebUI image, see Stable-Diffusion-WebUI.